Added two-photon absorption nonlinearity

[caseyflex]

• Adds TwoPhotonAbsorption and NonlinearModel classes. Also, nonlinearities of different types can be combined. Note that the beta and n2 parameters for these classes can be complex. Also, these classes need n0, which can be provided or obtained automatically from the medium and source frequencies.
• Adds warning for normalizing against CustomSourceTime or ContinuousWave sources

[tylerflex]

just a few minor comments as I looked over this.

[tylerflex]

Looks good! just a few minor comments.

[momchil-flex]

I am a tad unhappy with the naming of numiters as opposed to num_iters. Generally we follow the num_*** convention in everything (e.g. num_poles, num_layers... and even the num_iters in the fitter). However, we already released the first iteration of nonlinear stuff in 2.4, so I don't know if we want to make that change. That said however, are you introducing any breaking changes here anyway?

[caseyflex]

I am a tad unhappy with the naming of numiters as opposed to num_iters. Generally we follow the num_*** convention in everything (e.g. num_poles, num_layers... and even the num_iters in the fitter). However, we already released the first iteration of nonlinear stuff in 2.4, so I don't know if we want to make that change. That said however, are you introducing any breaking changes here anyway?

I am not introducing any breaking changes here. I can rename numiters to num_iters, and keep the old parameter in NonlinearSusceptibility and make it still work for backward compatibility.

@tylerflex I would like to move num_iters into NonlinearSpec, rather than duplicating it in every model. Would you be ok with this? I know we didn't want too many specific parameters in the general NonlinearSpec, but I feel this one is sufficiently fundamental to move there. And that would save the user from having to type it in each NonlinearModel.

[tylerflex]

@tylerflex I would like to move num_iters into NonlinearSpec, rather than duplicating it in every model. Would you be ok with this? I know we didn't want too many specific parameters in the general NonlinearSpec, but I feel this one is sufficiently fundamental to move there. And that would save the user from having to type it in each NonlinearModel.

As long as you think we can be pretty sure that this will be a field in every nonlinear model we add, specifically that future nonlinear models will not need their own num_iters. If it's just about backwards compatibility, we can easily handle this I think by having num_iters and numiters but adding a validator on numiters which asks users to use num_iters and then just sets it with the supplied value. Also alias might be worth looking into. We might be able to set a field num_iters with an alias of numiters.

https://docs.pydantic.dev/latest/concepts/alias/

[caseyflex]

@tylerflex I added KerrNonlinearity as well as a note about how that relates to chi3

[tylerflex]

Thanks @caseyflex, pretty minor comments. Still a little confused about the numiters behavior so asked for some clarification (and potentially a strange behavior spotted in a validator), but otherwise looks good.